Mitochondrial Genomic Alterations in Pediatric Acute Lymphoblastic Leukemia: Implications for Methotrexate‐Induced Neurotoxicity and Ethnic Disparities

ABSTRACT

Methotrexate (MTX)‐induced neurotoxicity is a serious complication in pediatric acute lymphoblastic leukemia (ALL), with observed ethnic disparities in risk. To investigate the role of mitochondrial DNA (mtDNA) variation, we performed whole mtDNA sequencing in 94 children with ALL, including 62 Hispanic patients. We identified 1294 mtDNA variants, including 12.4% novel variants, with MT‐ND5 and the D‐loop representing major variant hotspots. Gene‐based analyses identified significant associations of MT‐ND6 and MT‐ND2 with MTX neurotoxicity after multiple testing correction. Heteroplasmy was more frequent in affected patients, and predictive models incorporating mtDNA variants improved discrimination (AUC = 0.78) compared with clinical factors alone. These findings suggest that mtDNA variation, particularly involving Complex I genes, may contribute to MTX neurotoxicity and support its potential utility for risk stratification, while requiring validation in independent cohorts.

Acute lymphoblastic leukemia (ALL) is the most common cancer in children and a leading cause of childhood mortality in the US [1]. Treatment advances have substantially improved outcomes, with five‐year survival now exceeding 90%. Methotrexate (MTX) is a key component of combination chemotherapy for pediatric ALL. However, its use is often accompanied by severe neurotoxicity, including altered mental status, seizures, and stroke‐like symptoms. These complications not only significantly impact the quality of life but also contribute to decreased long‐term survival [2]. In a prior multi‐institutional prospective study of 280 pediatric ALL patients, we demonstrated that Hispanic children were significantly more likely than non‐Hispanic White children to develop MTX neurotoxicity, independent of established demographic and clinical risk factors, suggesting underlying biological heterogeneity contributing to this disparity [3]. Although genome‐wide association studies have identified over 19 loci associated with pediatric ALL and several linked to MTX neurotoxicity [4, 5, 6], the genetic underpinnings of MTX neurotoxicity across different ethnic groups in pediatric ALL patients remain poorly understood.

Mitochondria are essential organelles that regulate energy production, cell proliferation, and apoptosis. Given the high metabolic demands of neurons, mitochondrial dysfunction can lead to neuronal damage and death, contributing to neurotoxicity [7]. Emerging evidence implicates mitochondrial DNA (mtDNA) alterations in cancer development and treatment‐related side effects, including pediatric ALL ([8]). A recent analysis of pediatric B‐ALL and T‐ALL demonstrated increased somatic mtDNA deletion burdens and distinct deletion profiles across diagnosis, remission, and relapse, including large clonal deletions in some remission samples that may reflect treatment‐related selection or mitochondrial toxicity [9]. Whereas that study characterized predominantly somatic mtDNA alterations in leukemia samples, the potential association of germline mtDNA variation with MTX‐induced neurotoxicity remains largely unexplored. mtDNA variants also define maternal haplogroups, reflecting ancestry‐related geographic and population structure. In this study, we performed whole mtDNA genome sequencing on 94 pediatric ALL patients of different ethnic backgrounds. We aim to (1) characterize germline mtDNA variants in pediatric ALL; (2) identify mtDNA variants associated with MTX‐induced neurotoxicity; and (3) explore if mtDNA variants may contribute to ethnic disparities in MTX neurotoxicity.

Our study cohort consisted of 94 children with ALL, including 62 (66.0%) Hispanics. Informed consent was obtained from all participants. MTX‐induced neurotoxicity was defined as neurological symptoms, including stroke‐like episodes, aphasia, or seizures, occurring within 21 days after intrathecal or intravenous MTX administration and prompting modification of MTX therapy; each suspected case was independently evaluated by two pediatric oncologists. A total of forty patients (42.6%) developed MTX‐neurotoxicity during or after treatment (Table S1). The Whole mitochondrial genome from peripheral blood lymphocyte DNA was enriched by long‐range PCR and sequenced on the Illumina HiSeq 2000. Detailed methods are described in the Figure S1.

Across the cohort, we identified 1294 mtDNA variants (Figure S1), of which 161 (12.4%) have not been reported in public databases. Consistent with previous studies ([10, 11]), the mitochondrial control region (D‐Loop) harbored the highest proportion of variants (32.1%). A total of 689 variants were identified in the 13 protein‐coding genes, with MT‐ND5 (8.7%) showing the highest frequency, followed by MT‐CYB (7.6%) (Figure 1 and Figure S2). We also identified 111 MT‐rRNA variants (8.6%) and 69 MT‐tRNA variants (5.3%). Similar distributions were observed in Hispanic and non‐Hispanic patients (Figure S3). These findings are consistent with previous reports identifying MT‐ND5 as a mutational hotspot in multiple cancer types [11, 12] and highlight a diverse and partially novel landscape of mtDNA variation in pediatric ALL.

FIGURE 1.

The distributions of mtDNA variants in pediatric all patients by their biotype. This bar plot depicts the number and types of mitochondrial DNA (mtDNA) mutations observed in each mitochondrial gene and region. Variant types are color‐coded: RRNA (purple), missense mutations (orange), frameshift mutations (gray), tRNA mutations (yellow), nonsense mutations (magenta), control region variants (green), and non‐coding region variants (navy blue). Each gene or region along the x‐axis is associated with stacked bars representing the count of each mutation type.

The median number of mtDNA variants per patient was 53 (range: 22 to 416). A higher variant burden was associated with older age at diagnosis (p < 0.006) and higher BMI (p = 0.014), but not with MTX neurotoxicity risk (OR, 1.01; p = 0.36). Forty‐four mtDNA variants were predicted to be potentially pathogenic, including a known pathogenic tRNA variant (m.15923A>G) previously linked to mitochondrial diseases.

In single‐variant analyses of common variants (allele frequency > 5%), no associations remained significant after Bonferroni correction (p < 0.05/1294 = 3.86 × 10⁻⁵). Variants with nominal significance (p < 0.05) are listed in Table S2 and should be interpreted with caution and considered hypothesis‐generating pending validation in larger independent cohorts. In gene‐based analyses, using a Bonferroni‐corrected significance threshold (p < 0.05/16 = 3.125 × 10⁻³), MT‐ND6 was significantly associated with MTX neurotoxicity in the SKAT‐O test (p = 1.95 × 10⁻³), whereas MT‐ND2 was significant in the burden test (p = 9.05 × 10⁻⁴) (Table 1). Both genes encode subunits of mitochondrial Complex I. This convergence identifies Complex I as a candidate pathway for further investigation but does not demonstrate that these variants alter Complex I function or cause MTX‐induced neurotoxicity. Complex I is essential for oxidative phosphorylation (OXPHOS) and ATP production, and its dysfunction has been associated with increased oxidative stress, impaired mitochondrial function, and neurotoxicity [13, 14, 15, 16].

TABLE 1.

Associations between mitochondrial genes and MTX neurotoxicity.

Gene	BT	SKAT	SKAT‐O
Control region	0.22	0.00	0.01
Complex I
ND1	0.02	0.01	0.01
ND2	9.05 × 10 −4	0.01	0.00
ND3	0.01	0.01	0.01
ND4L	0.16	0.02	0.04
ND4	0.03	0.01	0.01
ND5	0.02	0.01	0.01
ND6	1.44 × 10 −3	4.30 × 10−3	1.95 × 10 −3
Complex III
CYB	0.04	0.01	0.01
Complex IV
CO1	0.04	0.01	0.02
CO2	0.01	0.02	0.01
CO3	0.03	0.02	0.03
Complex V
ATP6	0.06	0.01	0.02
ATP8	0.11	0.05	0.09
tRNA	0.03	0.01	0.02
rRNA	0.02	0.02	0.02

Note: Models are adjusted for age at diagnosis, BMI Z‐score, sex, treatment arm, and the first 10 principal components. Bold represents significant p values after Bonferroni correction.

Abbreviations: BT, burden test; SKAT, Sequence Kernel Association Test; SKAT‐O, optimized sequence kernel association test.

Our previous study showed that Hispanic children with ALL were more likely than non‐Hispanic White children to develop MTX neurotoxicity [3]. To explore whether mtDNA variation contributes to this disparity, we first analyzed mtDNA haplogroups. Fourteen haplogroups were identified, with haplogroup A being most common in Hispanic children, consistent with its known prevalence in populations with Native American ancestry. However, no association was observed between haplogroups and MTX neurotoxicity (Table S3). In ethnicity‐stratified analyses, the control region and MT‐ND2, MT‐CO2, MT‐ND6, and MT‐CYB genes showed nominal associations with MTX neurotoxicity among Hispanic children (p < 0.05), but not among non‐Hispanic children (Table S4). Haplogroup distributions also differed by ethnicity (Table S5). Given the limited sample size, these findings are exploratory and may reflect population genetic structure or stochastic variation rather than true ancestry‐related biological differences. Validation in larger, independent cohorts is required to clarify their relevance.

Because mitochondrial heteroplasmy has been implicated in cancer biology and mitochondrial dysfunction, we evaluated its association with MTX neurotoxicity. Among the 94 pediatric patients, 51 (54.3%) patients harbored heteroplasmy (Figure S4A). Patients who developed MTX neurotoxicity were more likely to exhibit mitochondrial heteroplasmy than those without (70.0% vs. 42.6%, p = 0.015, Figure S4B).

We next developed predictive models using demographic parameters (age at diagnosis, BMI z‐score, sex and ethnicity), clinical factors (diagnostic type and treatment risk arm), and mtDNA variants, individually and in combination (Table S6). Given the limited sample size, separate training and testing datasets could not be generated; therefore, ten‐fold cross‐validation and bootstrap resampling were used for internal validation. In the cross‐validation analysis, models based on demographic, clinical, and mtDNA variables alone achieved AUCs of 0.54, 0.75, and 0.65, respectively, for predicting MTX neurotoxicity (Figure S5). The integrated model achieved an AUC of 0.78, outperforming the model based solely on demographic and clinical factors. Similar results were observed using bootstrap validation. The predictive performance of the demographic and clinical model was comparable to that reported in our previous study using 97 pharmacogenomic variants [17]. Incorporation of mtDNA variants improved model discrimination and potential utility for risk stratification. However, because the reported performance may be optimistic in the absence of an independent validation cohort, the model should be considered a proof of concept rather than a clinically deployable prediction tool. Independent validation in larger cohorts will be required to establish its generalizability and clinical utility.

Our study has several limitations. First, although we focused on germline mtDNA variants, determining their origin, specifically distinguishing inherited variants from potential somatic events, would require mtDNA sequencing from maternal or other family members. Such samples are not currently available for this cohort. Second, this study used targeted whole‐mtDNA sequencing rather than whole‐genome sequencing; therefore, nuclear germline variants associated with MTX pharmacology or neurotoxicity were not comprehensively evaluated. The observed mtDNA associations may consequently be confounded or modified by unmeasured nuclear genetic variation and should not be interpreted as independent mtDNA effects. Third, we cannot determine whether the variants enriched in Hispanic children are functional variation that impacts the risk of MTX‐induced neurotoxicity. The observed associations do not demonstrate a mechanistic or causal relationship. Functional studies, confounder‐adjusted analyses incorporating nuclear genetic variation, and independent replication will be required to determine the biological significance of the implicated variants. Fourth, because mtDNA variants are correlated, Bonferroni correction may be overly conservative. Although gene‐based SKAT‐O and burden tests reduce the testing burden, this small exploratory study remains underpowered to detect corrected single‐variant associations. The findings should therefore be considered hypothesis‐generating and require validation using haplogroup‐aware or penalized methods in larger cohorts. In addition, although external validation is the gold standard for assessing model transportability and generalizability, we were unable to externally validate our risk prediction models because paired clinical and mtDNA sequencing data with MTX neurotoxicity outcomes are not yet widely available. The limited number of pediatric ALL cohorts with both mtDNA sequencing and MTX neurotoxicity phenotyping currently precludes independent replication, and additional time will be required for sufficient data accrual. We therefore relied on ten‐fold cross‐validation and bootstrap resampling for internal validation and observed consistent results across approaches. However, this internal validation framework cannot fully exclude overfitting, and the model should be interpreted strictly as a proof‐of‐concept for the integration of multi‐omic features rather than a clinically deployable prediction tool. Future studies in larger and more ethnically diverse cohorts, with independent external datasets, will be required to establish true generalizability and clinical utility.

In conclusion, this focused exploratory study identifies associations between mtDNA variation and MTX‐induced neurotoxicity in pediatric ALL. These findings do not establish causality or clinical utility but highlight mitochondrial genomic variation as a candidate for investigation in larger, independently validated, and functionally characterized studies.

Author Contributions

Paul L. Auer: writing – review and editing, formal analysis. Raul Urrutia: writing – review and editing, formal analysis. Jing Dong: conceptualization, funding acquisition, writing – review and editing, methodology, formal analysis, project administration, resources, supervision, data curation, investigation, writing – original draft. Wael Saber: writing – review and editing, formal analysis. Laura Michaelis: writing – review and editing. Xiaowu Gai: writing – review and editing, methodology. Philip Lupo: writing – review and editing, conceptualization, writing – original draft, resources, formal analysis, data curation. Michael Scheurer: resources, writing – review and editing, conceptualization, funding acquisition, writing – original draft, formal analysis, data curation. Shahram Arsang‐Jang: writing – review and editing, methodology, formal analysis. Jing Han: writing – review and editing, writing – original draft, formal analysis, methodology.

Funding

This work was supported by Cancer Prevention and Research Institute of Texas, RP160097, RP160771. National Heart, Lung, and Blood Institute, K01 HL164972.

Ethics Statement

The current study was approved by the Institutional Review Board of Texas Children's Cancer Center and the Medical College of Wisconsin and conducted in accordance with the Declaration of Helsinki.

Consent

Informed consent was obtained from all subjects involved in the study (IRB# H‐29892).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Mitochondrial Genomic Landscape in Pediatric ALL patients. This circular visualization illustrates the distribution and types of variants across the entire mitochondrial genome. Outer colored bands indicate gene regions, including protein‐coding genes (green), rRNA genes (blue), tRNA genes (pink), intergenic regions (white), and the control region (orange). Dots in the inner ring represent individual variant, with each color corresponding to a specific nucleotide change as indicated in the mutation type legend. X‐axes represent the heteroplasmic level of each variant, ranging from 0 to 1.

Figure S2: Distribution of mtDNA variants across mitochondrial regions.

This dot plot illustrates the number of mtDNA variants observed in individual samples across various mitochondrial genomic regions. Each dot represents the variant count for a specific sample in each gene or region. Regions include the control region, tRNA genes, rRNA genes (RNR1 and RNR2), and protein‐coding genes (ND1, ND2, ND3, ND4, ND4L, ND5, ND6, CYTB, CO1, CO2, CO3, ATP6 and ATP8).

Figure S3: The distribution of mtDNA variant across mitochondrial genes and regions by ethnicity. This figure presents the number and classification of mtDNA variants detected in each mitochondrial gene and region by ethnicity.

Figure S4: mtDNA heteroplasmy and associations with MTX neurotoxicity. (A) Histogram showing the distribution of mitochondrial heteroplasmy counts per patient across the cohort. (B) Bar plot comparing the number of individuals with or without heteroplasmy, stratified by the presence or absence of MTX neurotoxicity.

Figure S5: Receiver Operating Characteristic (ROC) curves for models predicting MTX‐induced neurotoxicity in the cross‐validation methods.

Table S1: Clinical and demographic characteristics of patients.

Table S2: Associations between mtDNA variants and the risk of MTX neurotoxicity at p < 0.05.

Table S3: Associations between mitochondrial haplogroups and the risk of MTX neurotoxicity.

Table S4: Stratification analysis by ethnicity.

Table S5: The distribution of haplogroups in different ethnic groups.

Table S6: The AUCs of the predicting models.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.